How do the IR-sensors work in sloping terrain?

[martinpi]

Sorry if it has been answered before. At least it has been asked before:

What happens to an IR-sensor stabilized plane in sloping terrain, i.e. near a hill, mountain or building?

Let me guess:
With an autopilot of the ArduPilot type (see http://www.diydrones.com: sensors control ailerons and elevator, autopilot controls rudder and throttle), assuming that the hill is on my left side:
The IR sensors raise the left wing, causing the plane to go right. The autopilot notices a course deviation and tries to compensate with the rudder. The effect is a slip maneuver and this will cause a loss of altitude. The Autopilot will try to compensate with the throttle.
A slight slope would result in a strange flight pattern, a steeper slope would cause a crash.

With the Paparazzi autopilot, controlling the ailerons, no rudder (FunJet or similar):
The plane would start to roll to the right as above, causing it to go right. The autopilot, still thinking that the plane is level, will fly a curve to the left by calculating a setpoint for the roll-angle. This should bring the plane back to level flight, although the autopilot thinks that it flies with its left wing down.

Is this correct?
In the Paparazzi simulator I tried to change the IR bias. This resulted in flying slightly right or left of the course or in flying in a circle when overdone.

I did not yet do the puzzle for flying towards the hill or away from it.

[dmgoedde]

I have considered this topic carefully for a long time, and have reason to believe I am an expert on it. I would like to see what the Paparazzi guys have to say on it. I have absolutely no idea how the Paparazzi system handles this, or if it even does. I have never reviewed any of their code. My development path on this is from scratch.

Because these types of stabilization systems have no IMU, they have no idea of level except for what it can "see" via the thermopiles. If the horizon is skewed, the plane will attain what it sees to be level attitude, however if this does not match gravity level, plane will bank and turn and/or change altitude.

The way I developed the thermopile version of AttoPilot, it constantly compares apparent horizon from thermopiles against an accurate barometric climb rate and GPS heading turn rate. The errors are used to constantly update a bias for pitch and roll. The raw thermopile pitch and roll are corrected with this bias. Therefore, the plane knows what true level is despite even severe horizon disturbances, and changing terrain as plane flies along.

Also, you need to keep in mind thermopiles has a VERY wide view angle, approx 110 degrees. They don't get thrown off by a single mountain peak in the distance, or a single tall building.

[dentompie]

When my autopilot was still using thermophiles, I used a scheme similar to Dean's. I never tried it next to a mountain (there aren't any in Belgium ;-) ), but the problematic issues (change in ground texture would change the bias) were solved.
The only reason I switched to gyros and accelerometers was for the challenge of doing it. It's also more elegant once it works.

Dean,

Did you do any tests in foggy weather (or slightly foggy but 200meter high)? How accurate was the attitude estimate?

Tom

[dmgoedde]

Dean,
Did you do any tests in foggy weather (or slightly foggy but 200meter high)? How accurate was the attitude estimate?
TomWe rarely have fog (ever?) in Phoenix Arizona. Based on the characteristics of the sensors, I don't think the results would be suprising. Water vapor (fog) at say 200m altitude would appear to be the "sky" at long wave IR frequency. The air and fog at only 200m wouldn't be measurably cooler then the ground. I just don't see how it would work at all. I mentioed it before in this thread that an Atto beta tester flew into the bottom of a low cloud, and stabilization was compromised.